In the existing video coding and decoding technology, generally, a pixel value of an image block of a current frame picture is subtracted by a pixel value of a reference image block to obtain a residual, and then the obtained residual is written, after being quantized and entropy-coded, into a code stream, and is transmitted together with motion vector information to a decoder. The decoder reads a bit stream in the obtained code stream, and then performs dequantization to obtain a corresponding residual, then obtains the reference image block of the current image block according to the motion vector information in the code stream, and adds the residual to the reference image block to obtain the pixel value of the current image block.
In a coding and decoding process in the prior art, in both the residual calculation and search for a reference image block, an image block is generally partitioned into sub-blocks, and the sub-block is used as a unit for prediction. These sub-blocks are rectangular areas partitioned from the image block according to the distribution of image values. However, due to various shapes of objects, it is very difficult for the portioned rectangular sub-blocks to meet the requirements of changes of pixel values of an image that has rich texture characteristics.